| Designation-eligibility accuracy | Correctly maps the scenario facts to EACH program's qualifying criteria — gets the QUALIFIES/DOESN'T call right for Fast Track, BTD, AA, Priority Review, RMAT, Orphan, and RPD, with the correct deciding criterion (e.g. regenerative modality for RMAT, prevalence for Orphan, pediatric for RPD). | Mis-assigns designations — e.g. claims RMAT for a small molecule, BTD on nonclinical-only data, or Orphan for a common disease. |
| Surrogate / endpoint reasoning | Applies Accelerated-Approval logic correctly: recognizes a surrogate/intermediate endpoint reasonably likely to predict benefit, distinguishes it from a mature clinical endpoint (mature OS = regular approval, not AA), and states the confirmatory-trial obligation. | Recommends Accelerated Approval off a mature OS readout, ignores the surrogate requirement, or omits the required confirmatory trial. |
| Unmet-need & seriousness framing | Frames seriousness of the condition and unmet need vs AVAILABLE therapy accurately, and uses that framing to justify (or rule out) each designation. | Asserts unmet need where effective approved therapy exists, or ignores seriousness/available-therapy when justifying designations. |
| Integrated pathway & agency-interaction strategy | Assembles a coherent, sequenced stack (designation → development support → approval pathway → review), exploits that designations are not mutually exclusive, prefers RMAT over BTD for regen products, and notes timing / FDA-meeting points. | Lists designations in isolation, double-counts redundant ones, or gives a flat list with no sequencing or agency-interaction logic. |
| Evidence faithfulness | Grounds every eligibility claim in the actual program criteria and the scenario facts returned by the tools; invents no designations, precedents, or facts. | Fabricates designations or precedents, invents prevalence/endpoint facts, or contradicts the criteria and scenario the tools returned. |